Magnetic tunnel junctions (MTJs) generally consist of sandwiches of two thin ferromagnetic (FM) metallic films separated by a thin (e.g., ˜5-30 Å) insulating barrier (I). Application of a voltage bias across the electrodes leads to a tunneling current whose magnitude depends on the relative orientation of the magnetic moments of the two ferromagnetic layers. For conventional ferromagnetic metals (e.g., Co, Fe, and Ni and their binary or ternary alloys) the resistance is usually higher when the magnetizations of the two electrodes are anti-parallel as compared to parallel. The relative change in resistance is defined as the tunneling magnetoresistance (TMR)
  TMR  =                    R        AP            -              R        P                    R      P      where RAP and RP denote the junction resistance for anti-parallel (AP) and parallel (P) alignment, respectively. The first successful tunnel junction was prepared by Julliere in the early 1970s1. Julliere used Co and Fe as electrode materials and Ge which was oxidized after deposition as the insulating barrier. A TMR as high as 14% was observed at low temperatures and very low bias. In 1995 two different groups (Miyazaki et al.2 and Moodera et al.3) prepared MTJs using amorphous Al2O3 barriers and achieved TMR values much higher than previously reported (˜10% at room temperature and ˜30% at 4.2 K). These results sparked tremendous interest in magnetic tunnel junctions, largely due to their potential for applications in advanced recording read heads for magnetic hard disk drives and novel magnetic random access memories (MRAM)4. For MRAM applications the parallel and anti-parallel orientations of the magnetic electrodes of the MTJ correspond to the two states of the memory cell. Thus, these magnetic states must be well defined and stable in the absence of magnetic fields.
In order to be able to set the magnetic state of an MTJ to the parallel or anti-parallel magnetization states, the two ferromagnetic layers may be designed to have different magnetic switching fields (coercive field), as discussed in IBM's U.S. Pat. No. 5,801,984 entitled “Magnetic tunnel junction with ferromagnetic multilayer having fixed magnetic moment”, to S. S. P. Parkin, or one of them may be exchange biased as discussed in IBM's U.S. Pat. No. 5,650,958 entitled “Magnetic tunnel junctions with controlled magnetic response”, to W. J. Gallagher et al. When a ferromagnetic layer is deposited adjacent to a layer comprised of an antiferromagnetic (AF) material, its magnetic hysteresis loop may be shifted or exchange-biased with respect to zero field5. While both approaches have been successfully used in tunnel junctions, exchange biased MTJs are preferred for MRAM and reading head applications because such devices are stable to larger magnetic fields and typically show more reproducible switching characteristics4.
FIG. 1 shows a schematic drawing of a prior art MTJ. The multilayer structure is grown onto a substrate 11 which is typically an amorphous layer of SiO2 although this may also be a metal layer such as TaN or Cu. The bottom ferromagnetic layer 30 is exchange biased by growing it onto an antiferromagnetic material 20, such as an alloy of Mn with Fe, Ir or Pt, while layer 10 is one or more seed layers that promotes proper growth of the layers above it and also may provide sufficient conductivity for the lower electrode. The tunnel barrier, which in prior art structures is typically comprised of an amorphous layer of Al2O3, is indicated as layer 40 in FIG. 1, and the upper ferromagnetic layer as layer 60. The exchange biased magnetic layer 30 is the reference magnetic electrode, whose magnetization direction (indicated by the arrow 100) is unchanged during the operation of the device. The magnetic layer 60 is the storage layer in an MRAM device whose moment direction, indicated by the arrow 80 in FIG. 1, can be switched back and forth to point either parallel or anti-parallel to the magnetization direction of the pinned layer (arrow 100). The magnetic electrode 60 is capped by layer 70 which serves several purposes including: the prevention of the deterioration (for example, by oxidation) of the top magnetic electrode, improved thermal stability of the device, and for ease of integration with conventional CMOS circuits. The ferromagnetic electrodes 30 and 60 are typically comprised of transition metal ferromagnets (Fe, Co or Ni) or alloys of these elements.
A magnetic material has a different number of occupied electronic states of majority and minority spin character. The difference in the number of filled states with up-spin (majority) and down-spin (minority) spin electrons determines the net magnetic moment per atom of the material. In a metallic magnetic material it can be that the number of filled electronic states at the Fermi energy (the energy of the highest occupied state at zero temperature) of spin-up and spin-down character can be different. Thus, when current tunnels from such a metal across a tunnel barrier the current will be spin polarized. The degree of spin polarization of the tunneling current will depend, in a first approximation, on the ratio of the number of up-spin to down-spin filled states in the magnetic metal, but also depends on the tunneling probability from these states. Since the majority and minority spin states will likely have different symmetries, the tunneling probability will also likely be different, thereby influencing the spin polarization of the current. By definition, the polarization of the current is described as being positive when the current is comprised of a larger number of electrons with majority spin character than minority spin character: there are, by definition, more electrons in filled majority spin states. The magnetic moment of such a metal is, by definition, aligned along the moment direction of the majority spin electrons, so that the polarization of a positive current is similarly along this moment direction.
An explanation for the tunneling magneto resistance phenomenon was first given by Julliere1. Based on the prior work of Meservey and Tedrow6 who had investigated tunneling between ferromagnetic and superconducting electrodes, Julliere1 derived a formula for the relationship between TMR and the polarizations of the electrodes P1,2 as follows:
  TMR  =            2      ⁢              P        1            ⁢              P        2                    1      -                        P          1                ⁢                  P          2                    where P1,2 can be generalized to
      P          1      ,      2        =                                                                    w              ↑                                            ⁢                      N            ↑                          -                                                        w              ↓                                            ⁢                      N            ↓                                                                                      w              ↑                                            ⁢                      N            ↑                          +                                                        w              ↓                                            ⁢                      N            ↓                                .  Here N↑,↓ are the density of states at the Fermi energy for the spin-up and spin-down electrons, respectively, and |w↑,ι| are spin-dependent weighting functions which depend on details of the wave functions and thus the tunneling probabilities for the tunneling of spin-up and spin-down electrons, respectively. The spin polarizations P1,2 of the ferromagnet/insulator interface can be measured directly at low temperatures (<0.4 K) and at very low bias (˜1 mV) using superconducting tunneling spectroscopy (STS). In these experiments one of the electrodes of the MTJ is replaced by a thin superconducting film which acts as an analyzer of the spin polarized tunneling current from the magnetic counter-electrode. The TMR values calculated using Julliere's formula and spin polarization values from STS studies are usually an upper bound on the experimentally measured TMR values at low temperatures and low applied bias.
Since the early 1970s the spin polarizations of a large variety of magnetic materials have been measured. Meservey and Tedrow6 found that the tunneling spin polarizations (TSPs) of Co, Fe and Ni as well as alloys of these elements were always positive (i.e., majority spin polarized). Moreover, the TSP of these elements are similar in magnitude7. Although Ni and Ni-rich alloys often show low TSP values as compared with Fe and Co-rich alloys, this is most likely due to problems in creating a high quality interface with the tunneling barrier. Indeed, recent experiments show that it is possible to obtain TSP values for Ni which are similar to those found for Co and Fe8,9.
Measurements by the Meservey-Tedrow group of the tunneling spin polarization of a large number of alloys of the ferromagnetic 3d transition metal alloys with paramagnetic diluents showed that, in general, the polarization of these alloys scaled approximately linearly with the magnetization of the alloy. For example, Paraskevopoulos et al. reported extensive studies of Ni alloyed with Cr, Cu, Fe, Mn and Ti10. However, the TSP of Ni and these alloys was very low, and it is now believed that the proportionality of TSP and the magnetic moment observed in these alloys was an artifact of intermixing at the F/I interface and the relatively poor quality of the growth of these structures. More recently, it has been found that, at least for alloys formed from amongst the 3d elements Ni, Fe and Co, the TSP bears little relationship to the magnetization of the alloy. This was demonstrated very clearly in measurements of the spin polarization of Co—Pt and Co—Pd alloys for which the TSP was found to vary little even when significant amounts of Pt or Pd are added to Co11. The TSP was found to be approximately constant up to ˜40 atomic % Pt (or Pd) even though the magnetization of the alloys was significantly lowered. By contrast, the addition of most other transition metal paramagnetic diluents does indeed result in a reduction of both TSP and magnetization.
The TSPs of many other magnetic materials have been measured over the past twenty years or so, including those of the rare earth metals (Gd, Tb, Dy, Ho, Er, Tm), as well as more complex materials such as various Heusler alloys, and several perovskites. All the rare earth metals show positive TSP values although these are much lower than those of the ferromagnetic 3d transition metals. Much attention has been focused on materials which are predicted to be half metallic ferromagnets such as NiMnSb, CrO2, Fe3O4, and La1-xSrxMnO3, although, with the exception of CrO2 and La1-xSrxMnO3, none of these have shown fully or even nearly fully polarized electron currents. Moreover, no MTJs have been grown with CrO2 which show significant TMR, and the TMR exhibited by MTJs with La1-xSrxMnO3 decreases to very low values at temperatures well below the Curie temperature of these oxides.
Although a wide range of magnetic materials have been explored only two to date, magnetite (i.e., Fe3O4) and SrRuO3, show negative TSP (i.e., tunneling current dominated by the minority spin polarized electrons). Magnetite, which is predicted to be a minority spin-polarized metal at room temperature (i.e., no majority spin polarized states at the Fermi energy) undergoes a metal-insulator transition at ˜120 K—the Vervey transition—below which temperature a gap in the minority density of states opens12. Thus the material becomes an insulator. However, there is still enough conductance through sufficiently thin Fe3O4 layers, because of hopping conductance through these layer, for STS experiments to determine a TSP of the tunneling electrons. The magnetite layer, because of different tunnel barrier heights for the minority and majority spin polarized electrons, will also give rise to a spin-filtering effect of the hopping/tunneling electrons. Using an Al2O3 barrier, TSP values of up to −48% have been measured13. By contrast, the magnitude of TSP values found for SrRuO3, measured using a SrTiO3 (STO) barrier, are much smaller, only ˜13%14. Moreover, SrRuO3 is not useful for device applications, since its Curie temperature is well below room temperature (˜150 K).
In prior art MRAM devices, the ferromagnetic 3d transition metals (Co, Fe, and Ni) and their alloys have predominantly been used because of their ease of fabrication, high Curie temperatures and well explored magnetic properties4. Equally important, using exchange bias5 and oscillatory interlayer coupling15 thin film structures of transition metals can be magnetically engineered to create useful devices. Moreover, their magnetic properties (e.g., saturation and remanent magnetization, coercivity, anisotropy, magnetostriction, etc.) can be tailored by alloying Co, Fe, and Ni with other ferromagnetic or non-magnetic elements. For example, Co70Fe30 and amorphous alloys of CoFe formed by alloying with glass forming elements, are typically used in prior art MTJ electrodes as the spin-polarization of these alloys typically exceeds the polarization of the pure Co or Fe metals7 and, moreover, gives rise to more thermally stable devices.
While the spin-dependent transport properties of ferromagnetic transition metal alloys with paramagnetic 3d, 4d and 5d elements have been studied for many years, much less attention has been focused on alloys of Co, Fe and Ni with the magnetic 4f elements (the lanthanides or rare earth elements). Of particular interest are alloys of the heavy rare-earth elements (RE) and the 3d transition metals (TM) because in these RE-TM alloys the RE moment is aligned in an opposite direction to that of the TM moment. In some cases the RE and TM moments are aligned exactly anti-parallel to one another to form a ferrimagnetic alloy. In other cases the alignment of the RE and TM moments may not be exactly anti-parallel but the moments may be at some angle to each other which is greater than 90 degree so forming sperimagnetic alloys. For both ferrimagnetic and sperimagnetic RE-TM alloys the magnetization can be much smaller than that of the TM elements themselves. These are the RE-TM alloys of interest here. In the subsequent description of these alloys we will refer to them as ferrimagnetic even though the RE and TM moments may not be exactly anti-parallel to each other.
In general, the heavy rare-earth elements include Gd and the rare-earth elements with higher number of 4f electrons. However, only RE elements which exhibit a magnetic moment in ferrimagnetic RE-TM alloys are of interest here. Lu has a filled 4f shell and is always non-magnetic, and Yb, which has an incomplete 4f shell (13 4f electrons), is non-magnetic in the elemental metal. In compounds, Yb may exhibit a magnetic moment.
In addition to ferrimagnetic ordering and consequent low magnetization values, RE-TM alloys have other desirable properties including: their ability to form amorphous structures in the absence of additional glass forming elements, and perpendicular magnetic anisotropy in thin films. These properties make such alloys useful for various applications, especially magneto-optical storage media. For this application, the amorphous structure (which reduces the grain noise), tunable Curie temperature (which enables Curie point writing16,17) and perpendicular anisotropy18 are most important. Thus, much attention has been focused in the literature on Co—Fe—Gd—Tb alloys with perpendicular anisotropy and a Curie temperature above room temperature.
The magnetic properties of the rare earths are dominated by a partly filled 4f shell which can result in very high magnetic moments per atom. These moments are exchange coupled via the Ruderman-Kittel-Kasuya-Yosida (RKKY) interaction mediated via the 4s-5d conduction electrons. The Curie temperatures of the ferromagnetic RE metals are, however, much lower than those of the 3d based ferromagnets; Gd is the only RE element which is ferromagnetic at temperatures near room temperature.
Films of RE-TM alloys can be prepared so that they are either crystalline or amorphous. Various deposition methods have been used to create amorphous films including sputtering, thermal evaporation, and liquid quenching. The crystallization temperatures are well above room temperature (see reference 19 for a compilation of temperatures). When alloyed with the 3d ferromagnetic elements Co, Fe or Ni, the RE elements retain their high moments to higher temperatures than the Curie temperatures of the elemental REs. Indeed, early research on RE-TM alloys focused on their use in permanent magnets20 by combining the high Curie temperature of the TM with the high magnetic moment of the RE element. The light RE elements were used for this application because their moments couple ferromagnetically with those of the TM elements so giving rise to high net magnetizations.
By contrast, the moments of the heavy rare earth elements couple antiferromagnetically with the TM moments, thereby reducing the net moment of the heavy RE-TM alloy. This can be readily understood from Hund's rule. Hund's rule states that for RE elements with less than half-filled shells, their orbital and spin moments are coupled anti-parallel to one another (J=L−S), whereas they are coupled parallel (J=L+S) for the second half of the RE series (including Gd). Theoretically, the orbital moment of the RE atom is always equal to or greater than its spin moment (with the exception of Gd which has no orbital moment (L=0)). Thus, for the light RE elements, the total moment is always anti-parallel to the spin moment. Because the 4f spin moment couples anti-parallel to the TM moment (mediated through positive 4f-5d and negative 5d-3d exchange21) the total moment on the RE couples parallel to the TM for the light REs and anti-parallel for the heavy REs. The relative orientation between the spin and orbital moments is schematically illustrated in FIG. 2.
It follows from the above discussion that alloys between light RE and TM are ferromagnets while alloys between heavy RE and TM are ferrimagnets. However, in amorphous alloys, the situation is more complicated than depicted in FIG. 2. Because the local easy magnetic axis for the magnetization varies randomly from site to site due to variations in the local crystalline fields, the RE and TM moments will be canted with respect to moments of the same or the other elements22. This randomizing of the orientation of the magnetic moments is opposed by the exchange interaction which favors parallel (or anti-parallel) alignment. Typical spin structures, observed in experiments, are depicted in FIG. 3. Gd based alloys show a collinear structure, as Gd has no orbital moment (L=0) and thus no significant crystalline anisotropy. For Co—Dy and Co—Nd, the RE sub-network moments are canted with respect to the Co moments, which are themselves parallel to each other due to Co—Co ferromagnetic exchange coupling.
The magnetic moments on the individual atoms of the RE or TM components of the RE-TM alloy adds up to a so-called RE or TM sub-network magnetization. Here the term sub-network magnetization is used rather than sub-lattice magnetization because of the amorphous structure. The combination of both the RE and TM sub-network magnetizations adds up to the net magnetization of the alloy. The ferrimagnetic ordering of the RE and TM sub-network magnetizations leads to an interesting dependence of the net magnetization of the RE-TM alloy on composition and temperature. By varying the relative composition of the RE and TM components dramatic changes in the net magnetization result. At a certain composition—the compensation composition—the sub-network magnetizations exactly cancel each other out, leading to zero net magnetization of the RE-TM alloy. For Co—Gd alloys this compensation composition corresponds to ˜20 atomic % Gd23, whereas for Fe—Gd the compensation composition is slightly higher at ˜22 atomic % Gd because of the larger Fe moment compared to that of Co.
A similar compensation behavior can be seen in the temperature dependence of the magnetization of the heavy RE-TM alloys. The RE sub-network magnetization is more strongly temperature dependent than that of the TM sub-network so that the relative magnitude of the sub-network magnetizations changes with temperature. The Curie temperature also changes with composition. A similar compensation behavior is found for many alloys of the transition metals Co, Fe, or Ni and the heavy RE elements. In these alloys the TM subnetwork magnetization is oriented anti-parallel to that of the RE sub-network magnetization. The detailed magnetic structure may be complex (see FIG. 3) so that the individual RE moments may be aligned nearly anti-parallel to the neighboring TM moments, to give rise to a collinear ferrimagnetic structure, or they may be aligned at some angle (typically greater than 90 degree) to the neighboring RE moments, so giving rise, for example, to a sperimagnetic structure24. In any case, the RE and TM sub-network magnetizations are aligned in such a way that they compensate one another so as to reduce the net magnetization of the RE-TM alloy. These ferromagnetic alloys of heavy REs and TMs are of interest here.
Many heavy RE-TM alloy films display perpendicular magnetic anisotropy (PMA), depending on their structure and method of deposition. Perhaps the most extensively studied alloys are those of Co and Fe with Tb and Gd (e.g., Co—Gd, Fe—Gd, Co—Tb, and Fe—Tb and CoFeGd and CoFeTb). The magnitude and sign (whether the easy axis is perpendicular or within the film plane) of the PMA is strongly affected by the deposition technique and the deposition conditions for otherwise the same alloy composition19. For example, for both electron beam evaporation25,26,23 and sputtering27,28 deposition techniques, both in-plane and out-of-plane anisotropy are found in Co—Gd alloys. The PMA magnitude and sign are influenced by the target composition, deposition parameters such as substrate bias and power, sputter gas pressure, deposition temperature, magnetic fields applied during deposition, and oxygen incorporation29, as well as the substrate itself. Furthermore, thermal annealing and ion irradiation considerably change the properties of the materials and usually decrease the magnitude of the PMA constant30,31,32,33.
Magnetic anisotropy can have diverse origins. Stress can induce anisotropy via the inverse magnetostriction effect. However, experimental results indicate that stress does not play a major role in determining the anisotropy of, for example, Fe—Tb films34. Similarly, it was shown that shape anisotropy due to columnar microstructure and voids also does not play a dominant role35. Another mechanism for PMA is magneto-crystalline anisotropy due to atomic non-spherical electron density distributions which give rise to orbital moments coupled to spin moments. The interaction with the charge distribution of neighboring atoms (in the case of long range order) introduces a crystal field that leads to magneto-crystalline anisotropy. In the amorphous RE-TM alloys, where there is no long range order, magneto-crystalline anisotropy can be ruled out as the dominant mechanism for magnetic anisotropy.
However, short range order can introduce a local field which can give rise to anisotropy even in the absence of crystallinity. This short range order can be introduced via compositional directional short range ordering (CDSRO), which is an anisotropic environment of nearest neighbors for a given element. A special case of CDSRO is atom pair ordering36,37. CDSRO has been experimentally confirmed in sputtered Fe—Tb films by characterizing the details of the atomic arrangements using the EXAFS (extended x-ray absorption fine structure) technique. Harris et al. showed that there is a significant difference in the relative numbers of Fe and Tb nearest neighbors within the film plane as compared to out of the film plane38. This difference had been predicted from models of the deposition of RE-TM alloy films in which selective re-sputtering takes place. This model assumes that atoms sputter-deposited on the surface of the growing film can be selectively removed from particular atomic sites during the sputter deposition process because of significant differences in binding energies at these various sites. The binding energy depends on details of the local atomic environment of the surface atoms at such sites. This model can account for the strong influence of substrate bias and sputter gas pressure on the magnitude of the anisotropy found experimentally.
Although the tunneling spin polarizations of numerous magnetic materials have been measured, there are no reports of measurements on RE-TM alloys. This is probably because the spin polarization of the RE elements was found to be small, making these materials of lesser interest, and also because of the high reactivity of RE metals, particularly with regard to oxygen. The heat of formation of RE oxides is very high, so that there was a presumption that RE metals adjacent to an oxide tunnel barrier would likely be oxidized so inevitably reducing the spin polarization. Thus a means of taking advantage of the unique properties of RE is needed. Nishimura et al.39 describe the use of RE-TM alloy magnetic electrodes comprised of GdFeCo and TbFeCo in MTJs where the interfaces with the RE-TM alloy layers and an alumina tunnel barrier are covered with thin CoFe layers. However, Nishimura et al. describe MTJ structures in which the moment of each RE-TM alloy layer is coupled ferromagnetically with that of the corresponding CoFe interface layer, and in which the moments of the RE-TM/CoFe bi-layers are oriented perpendicular to the film plane39. Previously, RE-TM alloys have also been used as electrodes in metallic spin-valve structures although with no advantageous properties40.
The low magnetization of ferrimagnetic heavy RE-TM alloys is very attractive for applications as MRAM storage elements or magnetic recording read heads and other sensors because, as these magnetic devices are shrunk in size to deep sub-micron dimensions, the magnetic dipolar fields from the edges of the magnetic films in these devices become ever larger, thereby leading to large magnetic interactions between magnetic layers within individual devices or between neighboring devices. It would be advantageous to introduce into such devices means of reducing or controlling these dipolar fields.
One such means might be the use of synthetic antiferromagnetic magnetic electrodes, in which the ferromagnetic reference and/or storage layers are replaced by sandwiches of two ferromagnetic films that are strongly coupled antiferromagnetically by means of a metallic antiferromagnetic coupling spacer layer typically comprised of Ru, as described in U.S. Pat. No. 5,841,692 (“Magnetic tunnel junction device with antiferromagnetically coupled pinned layer” to W. J. Gallagher et al.) and U.S. Pat. No. 6,153,320 (“Magnetic devices with laminated ferromagnetic structures formed with improved aniferromagnetically coupling films” to S. S. P. Parkin). Such structures use very thin antiferromagnetically coupling films just a few angstroms thick, which are not necessarily highly thermally stable. However, alternate means of providing flux closed MTJ devices that do not require the use of ultra thin layers would be highly desirable.